Anthropic's move is quite outrageous; they secretly downgraded and only apologized after being caught. Once trust is broken, it's hard to rebuild.

CoinWorld reports that Anthropic was criticized by the community for secretly downgrading strategies, publicly apologized, and adjusted the Claude Fable 5 safety policy, removing silent downgrades. According to the terms, users are not allowed to use Claude to train competing products, and Anthropic plans to reduce the performance of suspected competing accounts without notice. Researchers warn that this move will interfere with third-party evaluations and open-source collaboration. Anthropic admits the mistake and will switch to a publicly visible warning system, expanding the screening scope in the future, which may inadvertently harm some harmless requests.